"Location-based Activity Recognition using Relational Markov Networks"

Review following a meeting with Nik N. The paper is by Lin Liao, Dieter Fox and Henry Kautz.

This paper used Relational Markov Models.
It created undirected clique templates which were instantiated to create an undirected belief network.
It used sampling techniques to find the mostly likely state given a set of observations.

The goal was to learn an activity. The observations assumed a distribution over a set of variables which were "NearX" variables like NearRestaurant, and NearStore. These observation then were used to determine activities which, in some cases, were highly correlated with a semantic label for a place. One activity per place was assumed.

The semantic labels were therefore coming from a hand-curated database, Microsoft MapPoint.